Fast Automated Unpacking and Classification of Malware
•Download as PPT, PDF•
4 likes•1,380 views
Recommended
Recommended
More Related Content
Similar to Fast Automated Unpacking and Classification of Malware
Similar to Fast Automated Unpacking and Classification of Malware (20)
More from Silvio Cesare
More from Silvio Cesare (16)
Recently uploaded
Recently uploaded (20)
Fast Automated Unpacking and Classification of Malware
- 1. Silvio Cesare Deakin University <silvio.cesare@gmail.com> Fast Automated Unpacking and Classification of Malware
- 2. Who am I and where did this talk come from? PhD student at Deakin University. DUPRA scholarship holder. Industry experience as a software architect. 50,000+ LOC committed to the Qualys scanner. Presented at CanSecWest, Blackhat, Ruxcon, and academic conferences. This talk is on my Masters research at CQUniversity.
- 3. Introduction
- 4. Background to the Study Malware is a significant problem. Used for criminal enterprise and financial gain. Static and dynamic approaches to malware analysis and detection. Traditional detection systems have used static string signatures. Static string signatures perform poorly with malware variants.
- 5. Aim and Scope of the Research Aim To efficiently and effectively detect and classify malware. Scope Windows executable programs only. Code unpacking, but not general deobfuscation. Static analysis for the classification.
- 6. Major Contributions An application level emulation based unpacker. Entropy analysis to detect end-of-unpacking. The use of a graph invariant to fingerprint control flow graphs. Decompiling control flow graphs to generate a signature. Set similarity search to show malware similarity. System implementation and evaluation.
- 7. Structure of the Presentation 1. Related Work 2. Problem Definition and Our Approach 3. Automated Unpacking 4. Malware Feature Extraction 5. Malware Classification 6. Conclusions and Future Work
- 8. Related Work
- 9. Malware Polymorphism • Syntactic Changes – Dead Code Insertion – Instruction Substitution – Variable Renaming – Code Reordering – Branch Obfuscation – Opaque Predication Insertion – Code Packing – etc • Semantic Changes – Code Insertion, Deletion, Substitution, Transposition
- 10. Malware Obfuscation Using Code Packing Traditional Code Packing Shifting Decode Frame Instruction Virtualization and Malware Emulators Resistance to Dynamic Analysis
- 11. Static Program Features • Object File Header Attributes • Bytes • Instructions • Basic Blocks • Control Flow Graphs • Call Graph • API Calls • Data Flow • Procedure Dependence Graphs • System Dependence Graph
- 12. Control Flow Features Figure . A control flow graph (left), and a call graph (right).
- 13. Comparison of Static Program Features Invariance under polymorphism. High level abstraction using program analysis. Robustness of feature extraction.
- 14. Classification of Program Features Vectors Strings Sets Graphs
- 15. Static Analysis of Malware Disassembly Control Flow Reconstruction Opaque Predicate Detection Alias Analysis of Assembly Language Obfuscation and Limits to Static Analysis
- 16. Automated Unpacking of Obfuscated Malware Detecting the Code Packing Transformation Dynamic Approaches Emulation and Virtualization Dynamic Binary Instrumentation Native Execution Hardware Paging Static Approaches Instruction Virtualization and Malware Emulators Not completely automated. Rotalume Semi-automated Static Approaches
- 17. Detecting End of Unpacking Renovo Pandora’s Bochs OmniUnpack Uncover Hump-and-dump
- 18. Static Approaches to Malware Classification Classification Approaches Statistical Classification Instance-Based Learning The Similarity Search Used in Instance-Based Learning Control Flow Based Classification Approaches Control Flow Graphs Call Graphs
- 19. Trends Malware Development Polymorphism Code packing using instruction virtualization and malware emulators Anti-analysis Static Malware Detection and Classification The use of program analysis techniques
- 20. Problem definition and our approach
- 21. Problem definition and our Approach Problem Definition Access to a database of malware Identify if a query is malicious by finding high similarity to an instance in the database. This is instance-based learning. Similarity is a real number [0,1]. Similarity greater than 0.6 indicates a variant. Our Approach Automated Unpacking Exact Flowgraph Matching Approximate Flowgraph Matching
- 23. Identifying Packed Binaries Using Entropy Analysis Entropy describes the amount ‘information’ in data. Compressed and encrypted content has high entropy. Typical packed malware is characterized as being compressed and encrypted.
- 24. Application Level Emulation Interpretation Windows API. Improvements to Emulation Emulating Known Sections of Code Verification of Emulation
- 25. Entropy Analysis to Detect End-of unpacking Standard approach is to exit unpacker when executing dynamically generated code. Doesn’t handle multiple layers. Solution is to check entropy of unread memory. Low entropy means no more data to unpack.
- 26. Discussion Instruction virtualization and malware emulators a problem. Anti-emulation a problem. Assumption that unpacking removes all significant obfuscations. Application level emulation good at working with many malware. Best at unpacking variants of known packers.
- 27. Evaluation Tested packing Windows programs hostname.exe (shown) and calc.exe prototype against 14 public packing tools. Results indicate accurate detection of the original entry point, and a speed suitable for adoption in real-time desktop Antivirus. Name Revealed code and data Number of stages to real OEP Stages unpacked % of instr. to real OEP unpacked upx 13107 1 1 100.00 rlpack 6947 1 1 100.00 mew 4808 1 1 100.00 fsg 12348 1 1 100.00 npack 10890 1 1 100.00 expressor 59212 1 1 100.00 packman 10313 2 1 99.99 pe compact 18039 4 3 99.98 acprotect 99900 46 39 98.81 winupack 41250 2 1 98.80 telock 3177 19 15 93.45 yoda's protector 3492 6 2 85.81 aspack 2453 6 1 43.41 pepsin err 23 err err Name Time (s) Num. Instr. mew 0.13 56042 fsg 0.13 58138 upx 0.11 61654 packman 0.13 123959 npack 0.14 129021 aspack 0.15 161183 pe compact 0.14 179664 expressor 0.20 620932 winupack 0.20 632056 yoda’s protector 0.15 659401 rlpack 0.18 916590 telock 0.20 1304163 acprotect 0.67 3347105 pespin 0.64 10482466
- 29. Static Analysis Built on a general platform for binary and program analysis. Disassembly Translation to intermediate language Control flow reconstruction Transformation of control flow to signatures
- 30. Exact Flowgraph Matching A graph invariant is invariant for isomorphic graphs. Useful as an estimate of graph isomorphism. Label nodes in graph using depth first ordering. List the nodes and their edges in order as a string. Hash the string to generate a signature. Assign a weight to the signature based on the number of basic blocks.
- 31. Approximate Flowgraph Matching Decompile control flow graphs to strings. Similar control flow graphs have similar strings. Edit distance between strings used to measure string similarity. Approximate dictionary search to index by distances uses metric trees. Assign a weight based on the length of the string.
- 32. The Approximate Flowgraph Signature Figure . The relationship between a control flow graph, a high level structured graph, and a signature.
- 33. Discussion Static analysis required to successfully disassemble, and reconstruct flow graphs. Requires unpacking to work else the packer may be classified. Assumes code packing is the primary form of obfuscation. In most cases, if unpacking is OK, good flow graphs can be extracted.
- 35. Malware Classification Using Set Similarity Construct a mapping between features of programs. Based on greedy matching. Asymmetric similarity is sum of product of similarities between mapped features and weights. Similarity is product of asymmetric similarities. ∑ ≥ < = i edxed ed x twweightw tw S iii i , ,0 di SSdiS =),(
- 36. The Set Similarity Search Find all sets from database that have a similarity > 0.6 to the query. Take advantage that many elements in set are fairly unique. Match these elements first. Cull sets. Then match frequently occurring elements.
- 37. Complexity Analysis Complexity of search is logarithmic to database size in expected case. If many duplicate or similar malware in database, complexity increases. Don’t store malware > 95% similar in database.
- 38. Implementation 100,000 LOC Mostly C++ SQL Malware Database Python/Bash scripts for statistics and evaluation Java GUI HTML/PHP Web Interface
- 39. Effectiveness of Approximate Matching Tested classifying Klez (left) and Roron (right) families of malware. a b c d g h a 0.84 1.00 0.76 0.47 0.47 b 0.84 0.84 0.87 0.46 0.46 c 1.00 0.84 0.76 0.47 0.47 d 0.76 0.87 0.76 0.46 0.45 g 0.47 0.46 0.47 0.46 0.83 h 0.47 0.46 0.47 0.45 0.83 ao b d e g k m q a ao 0.70 0.28 0.28 0.27 0.75 0.70 0.70 0.75 b 0.74 0.31 0.34 0.33 0.82 1.00 1.00 0.87 d 0.28 0.29 0.50 0.74 0.29 0.29 0.29 0.29 e 0.31 0.34 0.50 0.64 0.32 0.34 0.34 0.33 g 0.27 0.33 0.74 0.64 0.29 0.33 0.33 0.30 k 0.75 0.82 0.29 0.30 0.29 0.82 0.82 0.96 m 0.74 1.00 0.31 0.34 0.33 0.82 1.00 0.87 q 0.74 1.00 0.31 0.34 0.33 0.82 1.00 0.87 a 0.75 0.87 0.30 0.31 0.30 0.96 0.87 0.87
- 40. Effectiveness of Exact Matching Klez, Netksky, Roron (Clockwise) a b c d g h a 0.76 0.82 0.69 0.52 0.51 b 0.76 0.83 0.80 0.52 0.51 c 0.82 0.83 0.69 0.51 0.51 d 0.69 0.80 0.69 0.51 0.50 g 0.52 0.52 0.51 0.51 0.85 h 0.51 0.51 0.51 0.50 0.85 aa ac f j p t x y aa 0.74 0.59 0.67 0.49 0.72 0.50 0.83 ac 0.74 0.69 0.78 0.40 0.55 0.37 0.63 f 0.59 0.69 0.88 0.44 0.61 0.41 0.70 j 0.67 0.78 0.88 0.49 0.69 0.46 0.79 p 0.49 0.40 0.44 0.49 0.68 0.85 0.58 t 0.72 0.55 0.61 0.69 0.68 0.63 0.86 x 0.50 0.37 0.41 0.46 0.85 0.63 0.54 y 0.83 0.63 0.70 0.79 0.58 0.86 0.54 ao b d e g k m q a ao 0.44 0.28 0.27 0.28 0.55 0.44 0.44 0.47 b 0.44 0.27 0.27 0.27 0.51 1.00 1.00 0.58 d 0.28 0.27 0.48 0.56 0.27 0.27 0.27 0.27 e 0.27 0.27 0.48 0.59 0.27 0.27 0.27 0.27 g 0.28 0.27 0.56 0.59 0.27 0.27 0.27 0.27 k 0.55 0.51 0.27 0.27 0.27 0.51 0.51 0.75 m 0.44 1.00 0.27 0.27 0.27 0.51 1.00 0.58 q 0.44 1.00 0.27 0.27 0.27 0.51 1.00 0.58 a 0.47 0.58 0.27 0.27 0.27 0.75 0.58 0.58
- 41. Effectiveness of Exact Matching (cont) 15,409 malware. Collected from mwcollect Alliance honeypots. 94% were more than 95% similar to existing malware. 35% were 100% identical after unpacking.
- 42. Efficiency of Exact Matching 809 malware. 86% malware processed in under 1.3 seconds. 0.25s median time for benign samples. Figure . Malware processing time. Figure . Benign processing time.
- 43. Efficiency of Exact Matching with a Synthetic Database Classification time only. Logarithmic trend line. Figure . Scalability of classification.
- 44. Malwise’s Resilience to False Positives Approximate matching (left), Exact matching (right). cmd.exe calc.exe netsky.aa klez.a roron.ao cmd.exe 0.00 0.00 0.00 0.00 calc.exe 0.00 0.00 0.00 0.00 netsky.aa 0.00 0.00 0.19 0.08 klez.a 0.00 0.00 0.19 0.15 roron.ao 0.00 0.00 0.08 0.15 cmd.exe calc.exe netsky.aa klez.a roron.ao cmd.exe 0.00 0.00 0.00 calc.exe 0.00 0.00 0.00 0.00 netsky.aa 0.00 0.00 0.15 0.09 klez.a 0.00 0.15 0.13 roron.ao 0.00 0.00 0.09 0.13
- 45. Conclusions and Future Work
- 46. Future Work Submitted for publication a real-time approximate flowgraph matching system. The use of a more unpacking approaches (some work done), and packer classification (some work done). Ensemble learning approaches may better in practice than only flowgraph features. Evaluation with larger malware sets (1+ million). We will make a public web service next year.
- 47. Conclusions Thesis will be made public soon. Program analysis and flowgraph based classification can be performed in real or near real-time on the desktop. It is effective at identifying malware variants.
- 48. Questions?